This paper focuses on the logic of categorization of three Chinese « encyclopedias » (leishu 類書), and the influence the categories have on the texts they contain. The choices made by the compilers are in fact not always easy to understand, and are indeed often quite subjective, which leads to think that leishu may not always prove reliable when used for research on specific topics. In order to exemplify this idea, I present here the case of a short narrative which can be found in three different leishu, namely the Yiwen leiju, the Taiping guangji and the Taiping yulan. Despite the text's different versions being roughly the same, I argue that its meaning is modified by the way it is categorized in each of these books. By their choices, the compilers therefore influence our perception of the writings found in leishu, which are one of our most important sources concerning ancient Chinese literature.